Method and system for performing speech recognition based on best-word scoring of repeated speech attempts

1. A method for improving word classification performance of a speech recognition system allowing multiple speech attempts from a user, the method comprising: storing a predefined vocabulary of word models and non-word models, the word models containing speech representations of acceptable words and the non-word models containing speech representations of non-words; comparing each of the speech attempts to the word models and the non-word models to determine a plurality of best words and corresponding word scores and non-word scores for each of the speech attempts; determining at least one common best word from among all the speech attempts; determining if the at least one common best word is a highest-ranking best word based on the corresponding word scores for all speech attempts by the user; if so, classifying the multiple speech attempts as the at least one common best word if the at least one common best word is the highest-ranking best word for all the speech attempts; and if not, performing an objective test on each of the at least one common best word to classify the multiple speech attempts.

2. The method as recited in claim 1 wherein performing the objective test further comprising: determining an objective measure for each of the at least one common best word for each speech attempt comparing each of the objective measures to a predetermined threshold; and classifying the multiple speech attempts by the user based on the comparison if the at least one common best word is not the highest-ranking best word for all the speech attempts.

3. The method as recited in claim 2 wherein determining the objective measure includes determining a ratio between the word score and the non-word score for each of the at least one common best word.

4. The method as recited in claim 2 wherein classifying includes rejecting the speech attempt in response to the objective measure of any of the allowed multiple speech attempts failing to meet the predetermined threshold.

5. The method as recited in claim 2 wherein classifying further comprises: determining an average objective measure for each of the at least one common best word; and accepting the at least one common best word having a best average objective measure when the objective measure of the at least one common best word of all of the speech attempts meets the predetermined threshold.

6. The method as recited in claim 1 wherein comparing includes storing each of the at least one best word and corresponding word and non-word scores after each speech attempt by the user.

7. The method as recited in claim 1 wherein determining the at least one common best word includes determining when a number of speech attempts by the user exceeds a predetermined allowable number of attempts.

8. A system for improving word classification performance of a speech recognition system allowing multiple speech attempts by a user, the system comprising: a memory for storing a predefined vocabulary of word models and non-word models, the word models containing speech representations of acceptable words and the non-word models containing speech representations of non-words; means for comparing each of the speech attempts to the word models and the non-word models to determine a plurality of best words and corresponding word scores and non-word scores for each of the speech attempts; means for determining at least one common best word from among all the speech attempts; means for determining if the at least one common best word is a highest-ranking best word based on the corresponding word scores for all speech attempts by the user; means for classifying the multiple speech attempts as the at least one common best word if the at least one common best word is the highest-ranking best word for all the speech attempts; and means for performing an objective test on each of the at least one common best word if the at least one common best word is not the highest-ranking best word for all speech attempts.

9. The system as recited in claim 8 wherein the means for performing further comprising: means for determining an objective measure for each of the at least one common best word for each speech attempt by the user representing a confidence level of the corresponding word and non-word scores; and means for comparing each of the objective measures to a predetermined threshold; wherein the means for classifying is further operative to classify the multiple speech attempts by the user based on the comparison if the at least one common best word is not the highest ranking best word for all the speech attempts.

10. The system as recited in claim 9 wherein the means for determining the objective measure includes means for determining a ratio between the word score and the non-word score for each of the at least one common best word.

11. The system as recited in claim 9 wherein the means for classifying includes means for rejecting the speech attempt in response to the objective measure of any of the allowed multiple speech attempts failing to meet the predetermined threshold.

12. The system as recited in claim 9 wherein the means for classifying further comprises: means for determining an average objective measure for each of the at least one common best word; and means for accepting the at least one common best word having a best average objective measure when the objective measure of the at least one common best word of all of the speech attempts meets the predetermined threshold measure.

13. The system as recited in claim 8 wherein the means for comparing includes means for storing each of the at least one best word and corresponding word and non-word scores after each speech attempt by the user.

14. The system as recited in claim 8 wherein the means for determining the at least one common best word includes means for determining when a number of speech attempts by the user exceeds a predetermined allowable number of attempts.